BYOM (Bring your own model)

BYOM (Build your own model) feature allows you to create your own data model that you can then train based on your own unique needs.

This solution allows you to simplify creation of models and predictions as all your data is stored and analyzed in one application. Once you migrate your compensation data, you are connected to it in real time without a need to pull it from a different source every time you want to make an analysis.

Models you'll create in BYOM feature will be tailored to your company's situation and needs and will contribute to your decision making process and fair compensation of your payees.

In this article you will find information about:

• Accessing BYOM

• Creating a model in BYOM

• Managing model training in BYOM

• Prerequisites to use BYOM

Accessing BYOM

To access the BYOM page, proceed as follows:

1. Select Admin in the navigation bar on the left side of the page. The new panel will appear.

2. In the list on the panel find "BYOM" section and select Models.

You can now access all the stored BYOM models and create a new one.

The following figure illustrates the interface of the BYOM feature of the beqom application:

Access to BYOM

Creating a model in BYOM

To set up your model, proceed as follows:

1. From the Data Scope drop-down list, select the data scope you wish to use for your model.

This is a tip message.

For more information about creating a scope please, see:BQM106-SCOPE_API

2. Press the Refresh Data button to create a snapshot of the data. This will load the data according to the defined data scope.

3. Click on the New Model button. A pop-up window will appear, as illustrated in the following figure:

   '

   Model creation in BYOM

4. Enter the model's Name and Description. Here you may also change the scope in the drop-down list.

5. Click New Model. You are now moved to the Population Criteria section.

6. Click Add to select the population criteria to filter the training data and map the model to payees for prediction.

7. From the first drop-down list select a criterion, from the second drop-down list choose between in or not in options.

8. Third selection bar appears with available options below in form of tags. Select the tags you wish to add to your criteria.

9. Click Add to add another criterion or the icon to delete the ones you already created. If no matching models are found based on the criteria, the system will use the last trained model without criteria. The following figure illustrates the filled out criterion fields:

   

   Criteria in BYOM

10. Once the population is created, on the bottom right corner of the page, click Discard to delete your population, Save Population to save your work, or Save and Go to Training Mode to proceed to training mode.

Managing model training in BYOM

To begin your model training, proceed as follows:

1. In the Training Mode section select the Data Version from the drop-down list.

2. In the Model features use the search bar to find the feature you wish to add or click Add Features and select them from the drop-down list that will appear in the pop-up window. Then click Apply to add them to the model. Features are filtered according to the population criteria defined earlier. The following figure illustrates the feature selected for the training:

   

   Selected features

3. After selecting the features, at the bottom right corner, click the Run Model Training button to run the training or Exit Training Mode to discard your model. If the model has intervals defined and the data causes multicollinearity, the intervals will not be calculated.

4. Once the training is complete (this may take some time), a pop-up is displayed with the results, including "Population Count", "Model Features", and "Model Performance", as illustrated in the following figure:

   

   Training results

5. If you are satisfied with the training results, you can publish the model by clicking the Save and Replace button.

Prerequisites to use BYOM

Before starting the training process, you must:

• Ensure that your data is normalized. This is crucial for achieving accurate and reliable results. Then import the data you wish to use for training into the salary_prediction_training table.

• Create a data scope to define the subset of data that the model will be trained on.

This is a tip message.

For more information about creating a scope please, see:BQM106-SCOPE_API

Once you uploaded all the scopes you wish to base your models on, you are now ready to create your model.